In the course of the analysis we have seen that there are two methods used which greatly help the selection of facts; namely, the phonetic «laws,» which leave a free hand in choosing comparative material, and semantic extensions, which permit the inclusion of related ideas. Indeed, both methods are powerful tools when properly used, but the question is as to the method of their application, — How far may they be applied? Of course, the limits of practical application of the phonetic «laws» are strictly outlined in general treatises, but it is very often forgotten, so we shall remember them.
We shall use the instances of the languages here discussed. The correspondence of the initial Manchu f, to Goldi p, to some Northern Tungus x, h, and zero, found out of the observation of facts, is beyond any doubt. However, as shown, there are many «exceptions» in which Manchu words with the initial bilabial correspond to Goldi words with the initial x, h, and zero; also the Manchu words without the initial bilabial correspond to Goldi words with the initial p. The number of facts disturbing the «ideal» correspondence is so great that one cannot consider it as a «law,» as a law is understood in science. The facts cannot thus be explained and predicted by the reference to this «law,» for the simple reason that it does not exist as a «law.» Furthermore, when the facts are explained by these types of occurrences in addition to being increased with other hypotheses, the explanation may happen to be absolutely erroneous and misleading, as it actually happens with all cases when Tungus h is explained as the loss of the initial p. When such a «law» is practically applied to the establishment of parallels, it results in the production of an absolutely artificial picture of actually non-existing relations. Let us take another instance — the frequent correspondence, in Tungus, of ng, ngg, g, w, u, etc., which also greatly facilitates the finding of common words. The initial g of Manchu in many instances corresponds to ng of some Northern Tungus dialects, but not all Manchu g's correspond to the ng of the same Northern Tungus dialects. Owing to this, one cannot postulate that the word found in the Northern Tungus with the initial ng will be found in Manchu with the initial g, and vice versa. This type of alternation does exist, but it is not a law, which may have force when one goes from law to the particular cases for establishing original sounds. Some Tungus dialects have developed another peculiarity; namely, the alteration of the initial s into h and x, but some other dialects have jealously preserved the original s. However, some dialects are not so particular about these sounds and use both of them in parallel. Let us take a «sibilating» dialect and compare it with a «glottaling» one. When we need to find parallels, we can go straight by comparing the words with the initial s with that of the other dialect with the initial x or h and thus practically to expect such an occurrence in all cases. However, this regularity has no reverse power, — s will not correspond in all cases to x, x~h, for there may be a different origin of x~h, e.g., through «aspiration.» Yet if one compares a «sibilating» dialect with another one which is not definitely «sibilating» nor definitely «glottaling,» the principle of regularity observed cannot be applied. It is evident that every such correspondence of sounds may have scientific value when applied to the well-studied dialects, and it cannot be applied to the dialects where the alternation of sounds is not yet carefully studied. Of course, one may expect to find such a correspondence, but one cannot postulate it and hypothetically restore the stems for their further application as proof of anything. With the extension of principles of variations over other pairs of dialects, one may naturally come to some general tendencies cnaracteristic of groups of dialects. However, they do not become valid for other groups of dialects. Let us suppose that we have such a well-established alternation of sounds, and we meet with another alternating series — for example s→sh, where every s of one dialect becomes sh in another dialect; but if we compare the latter with a «glottaling» dialect not all sh's will correspond to x~h.
What is practically observed is that the alternation of consonants affects almost all consonants, but every group of dialects have certain phonetic peculiarities. In so far as one confines oneself to the groups of the same type of phonetic variations, the «law» may have value; but when one goes from known to unknown languages, the «law» loses its value altogether and in still greater degree than in the case of a single pair of alternating sounds, for the chance of error is greater as more numerous are the alternating sounds. Without speaking of principles regulating «laws» of phonetic changes, which at the present time are hypothetic and as such can never be used for «proving» anything, we have to state that we meet with a great variety of sound alteration. In the case of aspiration and bilabialization, a long list of sounds are involved in the process. Yet the same sounds are also found to be subject to other alterations under the influence, for example, of palatalization, voicing, etc. The process of alteration of almost all consonants into j and a further alteration of j into other sounds makes it theoretically true that every consonant in certain conditions may change into any othrr consona.it. So that if we extend the validity of «laws» good for one pair of languages (or dialects) over other languages, and if we extend the «law,» the latter will lose its scientific value altogether. The more numerous the dialects compared, the less sure the practical value of such «laws,» for what is practically observed in languages is that if we generalize «laws,» then the series of alternating sounds may be connected between themselves in such a manner that any consonant may change into any other consonant. Thus the practical application of phonetic regular changes may be admitted as valid only for the group of languages (or dialects) in which they have been found. So that the nature of phonetic changes is such that the narrower the application of the phonetic «laws,» the higher their practical value.
What is done by A. Sauvageot has nothing to do with the strict scientific method of using empiric principles of sound variations. First of all, he promotes cases of simply frequent occurrences into «laws» which methodologically cannot be justified; second, he uses his «laws» in a reverse sense, i.e., going from the «laws» to the facts, he «restores» the facts, collects them to suit the «laws»; third, he extends «laws» over the languages for which these «laws» have not been established. Such a methodology nowadays is never used in the studies into the Indo-European language. Why should it be applied to the «Ural-Altaic» languages?